Microsoft Azure AI Engineer Associate (AI-102) Certification Training

You can provision cognitive resources securely, isolate networks, and manage credentials using Managed Identities and Azure Key Vault.

1. Creating and consuming Azure AI resources via Portal and Azure CLI90 min
   • az cognitiveservices account create
   • pricing tiers S0 and F0
   • endpoint URLs and custom subdomains
   • primary and secondary access keys
   • API version query parameters
   • resource group scoping
2. Configuring Managed Identities and Azure Key Vault integration90 min
   • system-assigned vs user-assigned identities
   • SecretClient SDK initialization
   • Key Vault access policies and RBAC
   • secret versioning retrieval
   • DefaultAzureCredential token acquisition
   • managed identity authentication flow
3. Implementing API Key rotation, network isolation, and Private Endpoints90 min
   • virtual network rules and firewall configuration
   • az network private-endpoint create command
   • private DNS zone configuration
   • key rotation automation via Event Grid
   • public network access disable flags
   • subnet delegation settings
4. Monitoring resource usage, cost alerts, and diagnostic logging in Application Insights90 min
   • diagnostic settings configuration
   • Log Analytics workspace KQL queries
   • Azure Budgets and action groups
   • telemetry tracing initialization
   • metric alert rule thresholds
   • diagnostic log category mapping
5. Managing service limits, quotas, and region availability60 min
   • regional quota increase requests
   • tokens per minute (TPM) limits
   • concurrent request throttling
   • service-level agreements (SLAs)
   • redundancy configurations
   • geo-replication settings
6. Foundational security auditing and compliance posture checks60 min
   • Microsoft Defender for Cloud integration
   • secure score evaluation
   • policy compliance scanning
   • least privilege permission review
   • access review configurations
   • audit log retention policies

Hands-on project: Provision an Azure OpenAI resource using Terraform or Azure CLI, wire its keys to Azure Key Vault, and lock down access using a private endpoint in a dedicated VNet.

You can extract bounding boxes, read multi-line text from images, and train custom image classifiers using Azure AI Vision.

1. Analyzing images and video streams with Azure AI Vision v4.090 min
   • ImageAnalysisClient instantiation
   • visual features parameter selection
   • dense captions generation
   • smart cropping aspect ratios
   • image metadata extraction
   • synchronous vs asynchronous analysis
2. Extracting printed and handwritten text using Read API90 min
   • read operation status polling
   • operation-location response header
   • bounding polygon coordinate mapping
   • word confidence score thresholds
   • line-by-line text assembly
   • multilingual text extraction flags
3. Training custom image classification and object detection models90 min
   • Custom Vision portal project setup
   • tagging images with bounding boxes
   • iteration training execution
   • prediction endpoint integration
   • evaluation metrics precision and recall
   • exporting models for edge devices
4. Facial detection, verification, and ethical AI compliance considerations90 min
   • face attributes detection API
   • liveness detection parameters
   • responsible AI facial recognition guidelines
   • privacy standards and consent
   • face rectangle coordinate parsing
   • facial feature vector extraction
5. Handling image size limitations and preprocessing workflows60 min
   • payload size restrictions
   • OpenCV image resizing and compression
   • format conversion JPEG to PNG
   • DPI normalization for OCR
   • batch image preprocessing scripts
   • error handling for corrupted files
6. Integrating vision outputs into downstream business applications60 min
   • mapping OCR text to database schemas
   • storing visual metadata in Cosmos DB
   • triggering alerts on anomaly detection
   • exporting structured CSV reports
   • retry policies for transient vision errors
   • logging failed image analyses

Hands-on project: Build a Python command-line utility that processes a batch of store receipt images using the Read API, extracts line items, and exports structured JSON.

You can extract key-value pairs, tables, and unstructured text from complex multi-page invoices and financial reports.

1. Extracting layout and structured data with Azure AI Document Intelligence90 min
   • DocumentAnalysisClient initialization
   • prebuilt-invoice model routing
   • analyze_document method parameters
   • table extraction cell indexing
   • key-value pair confidence intervals
   • document selection mark parsing
2. Training custom extraction models for non-standard receipts and invoices90 min
   • Document Intelligence Studio workspace setup
   • labeled vs unlabeled training datasets
   • model schema JSON definition
   • training job execution and monitoring
   • model composition for multi-form types
   • testing custom models against test sets
3. Combining Document Intelligence with Python pipelines for automated data entry90 min
   • polling asynchronous analyze operations
   • JSON result parsing for line items
   • error handling for unreadable fields
   • database insertion mapping with SQLAlchemy
   • transaction rollback on validation failure
   • idempotent data processing patterns
4. Handling multi-page PDFs and complex tabular structures under high throughput90 min
   • page selection ranges parameter
   • concurrency limits and batching
   • batch processing script orchestration
   • storage blob trigger configuration
   • memory management for large documents
   • dead-letter queue for failed parses
5. Securing sensitive document data in transit and at rest60 min
   • customer-managed keys for storage
   • SAS token generation with short expiry
   • redacting PII from extracted fields
   • audit logging for document access
   • secure deletion of temporary files
   • encryption key rotation intervals
6. Optimizing extraction accuracy and handling low-quality scans60 min
   • deskewing and contrast adjustment
   • specifying locale parameters for numbers and dates
   • handling handwritten annotations on forms
   • cross-validating extracted totals with item sums
   • human-in-the-loop review queues
   • tuning confidence thresholds for auto-approval

Hands-on project: Deploy an Azure Function triggered by Blob Storage uploads that automatically parses incoming multi-page vendor invoices using Document Intelligence and pushes data to a PostgreSQL table.

You can analyze sentiment, extract entities, translate text, and filter toxic user inputs using Azure Language and Content Safety services.

1. Sentiment analysis, key phrase extraction, and named entity recognition (NER)90 min
   • TextAnalyticsClient configuration
   • opinion mining aspect-based analysis
   • language detection parameters
   • PiiEntityRecognition categories and redaction
   • confidence score evaluation
   • batch document analysis requests
2. Building custom text classification and conversational language understanding (CLU) models90 min
   • Language Studio project creation
   • intents and utterances schema design
   • entities annotation and patterns
   • model deployment endpoint configuration
   • testing utterances via REST API
   • version management for CLU models
3. Translating text and speech in real-time with Azure Translator90 min
   • translator REST API authentication
   • profanity marker configuration
   • transliteration script to romanized text
   • dictionary lookup and alignment
   • array translation payload formatting
   • handling unsupported language codes
4. Configuring Azure AI Content Safety to detect toxic, abusive, or harmful content90 min
   • text analysis API severity thresholds
   • hate speech and violence filters
   • custom blocklists management
   • multimodal content screening setup
   • actionable response categories
   • handling moderation HTTP error codes
5. Optimizing language service performance and payload batching60 min
   • document character limit management
   • asynchronous job submission for bulk text
   • connection pooling for high-volume requests
   • cancellation token implementation
   • client-side retry logic for network drops
   • latency benchmarking across regions
6. Integrating language intelligence into conversational logging60 min
   • storing sentiment trends in time-series stores
   • alerting on sudden sentiment drops
   • tagging support tickets with extracted entities
   • automated routing based on intent classification
   • compliance logging for moderated text
   • dashboard visualization of content safety metrics

Hands-on project: Implement a content moderation filter middleware in Python that screens incoming chat messages for hate speech and PII before forwarding them to downstream storage.

You can transcribe audio streams, synthesize natural neural voices, and implement speaker verification workflows.

1. Converting Speech-to-Text with custom acoustic and language models90 min
   • SpeechConfig initialization and subscription keys
   • AudioConfig input stream setup
   • SpeechRecognizer continuous recognition events
   • handling session started and stopped callbacks
   • audio format requirements WAV PCM
   • custom speech model endpoint binding
2. Text-to-Speech synthesis using custom neural voices90 min
   • SpeechSynthesizer configuration
   • SSML markup for voice styling and breaks
   • voice style and role parameters
   • audio output stream manipulation
   • viseme generation for lip-syncing animations
   • saving synthesized output to audio files
3. Real-time audio stream translation across multiple languages90 min
   • TranslationRecognizer initialization
   • target languages configuration array
   • speech synthesis output from translation
   • latency optimization in streaming sessions
   • handling overlapping speech inputs
   • disconnect and reconnect recovery patterns
4. Implementing speaker recognition and verification workflows90 min
   • speaker identification API setup
   • voice profile creation and enrollment
   • audio sample length requirements
   • verification confidence score thresholds
   • text-dependent vs text-independent verification
   • security auditing of voice enrollments
5. Managing long-running batch transcription jobs90 min
   • transcription client REST API submission
   • blob storage container URI input binding
   • polling batch job status updates
   • downloading diarized JSON output files
   • handling failed batch audio segments
   • cost management for bulk transcription
6. Audio engineering best practices and noise cancellation30 min
   • sample rate and bit depth standards
   • noise suppression configuration flags
   • channel mixing stereo to mono
   • testing with simulated low-bandwidth audio

Hands-on project: Build an audio transcription script that reads long WAV meeting recordings, generates speaker-separated transcripts, and saves structured summaries.

You can build enterprise search indexes featuring vector embeddings, semantic reranking, and automated skillsets.

1. Creating and populating indexes with text, skillsets, and vector embeddings90 min
   • SearchIndexClient and SearchClient setup
   • edfields schema definition and data types
   • vector search profile and algorithm configuration
   • vectorizer configuration for embeddings
   • pushing document batches via SDK
   • handling merge and upload concurrency
2. Building custom AI enrichment pipelines using Azure Functions90 min
   • custom web API skillset integration
   • data enrichment pipeline lifecycle
   • indexer schedule and change tracking
   • push vs pull data source architectures
   • debugging skillset execution errors
   • passing context across skill nodes
3. Vector search, hybrid search, and semantic reranking implementations90 min
   • KNN (K-Nearest Neighbors) query syntax
   • semantic configuration and query options
   • reranker scoring and caption extraction
   • hybrid query combining keyword and vector scores
   • semantic answers and highlights parsing
   • tuning query parameters for precision
4. Managing index refresh schedules, aliases, and query performance tuning90 min
   • index aliases for zero-downtime reindexing
   • scoring profiles and custom weights
   • query filters and OData syntax
   • performance optimization and replica scaling
   • index backup and restore strategies
   • monitoring search latency metrics
5. Configuring security filters and document-level access control60 min
   • per-document security trimming fields
   • dynamic filter injection based on user roles
   • encrypting search service indexes
   • managing API keys for search clients
   • auditing search query logs
   • restricting access via IP firewalls
6. Handling incremental updates and data synchronization60 min
   • soft deletes in search indexes
   • high water mark column tracking
   • indexer error handling policies
   • resetting indexers and clearing caches
   • scheduling off-peak reindexing runs
   • validating index synchronization completeness

Hands-on project: Configure an Azure AI Search index over a collection of technical manuals, enabling hybrid search and semantic reranking with Python ingestion scripts.

You can deploy LLMs, engineer prompts, implement RAG architectures, and handle rate-limiting thresholds in production.

1. Deploying GPT-4, GPT-3.5-Turbo, and Embeddings models in Azure OpenAI90 min
   • Azure OpenAI Studio resource deployment
   • model version selection and upgrade paths
   • tokenizers and token counting libraries
   • API versioning header requirements
   • provisioned throughput deployment units
   • managing model capacity limits
2. Prompt engineering techniques, system message design, and parameter tuning90 min
   • temperature and top_p parameter trade-offs
   • system message construction for role enforcement
   • few-shot example formatting in prompts
   • structured JSON outputs using response format
   • chain-of-thought prompting patterns
   • mitigating prompt injection vulnerabilities
3. Implementing RAG (Retrieval-Augmented Generation) using Azure AI Search90 min
   • Azure OpenAI built-in data sources integration
   • LangChain and LlamaIndex integration with Azure endpoints
   • embeddings generation pipeline for source documents
   • context window injection and management
   • citation generation and verification
   • handling out-of-domain queries gracefully
4. Managing TPM (Tokens Per Minute) limits and building retry handling logic90 min
   • HTTP 429 rate limit error handling
   • exponential backoff with jitter implementation
   • rate limiting strategies across distributed callers
   • asynchronous request throttling in Python
   • load balancing across multiple Azure OpenAI instances
   • monitoring token consumption alerts
5. Fine-tuning custom models on Azure OpenAI60 min
   • preparing JSONL training datasets
   • uploading training files via python SDK
   • initiating fine-tuning job parameters
   • monitoring training and validation loss curves
   • deploying custom fine-tuned model weights
   • evaluating fine-tuned model performance
6. Function calling and tool use with Azure OpenAI60 min
   • defining tool schemas in JSON
   • handling tool call requests from model output
   • executing local functions and returning results
   • multi-turn conversation loops with tool execution
   • error handling during tool execution
   • securing function endpoints called by LLMs

Hands-on project: Build a RAG query application using Azure OpenAI GPT-4 and Azure AI Search that answers technical support questions referencing private company PDFs without hallucination.

You can design and deploy intelligent bots integrated with Azure OpenAI and Language understanding models.

1. Designing bots with Azure Bot Framework SDK and Composer90 min
   • ActivityHandler event routing in Python/C#
   • turn context management and state properties
   • dialog sets and waterfall dialog flow
   • adaptive cards JSON schema design
   • user prompt validation patterns
   • handling user interruption during dialogs
2. Integrating bots with CLU and Azure OpenAI for intelligent responses90 min
   • cognitive model dispatch across intents
   • prompt chaining inside bot message handlers
   • context persistence across conversation turns
   • state management with Cosmos DB storage
   • handling ambiguous user inputs
   • streaming LLM responses back to chat channels
3. Deploying bots to Web Chat, Teams, and custom mobile applications90 min
   • Azure Bot Service channel registration
   • Direct Line API token generation and security
   • Teams app manifest and bot registration
   • authentication tokens and OAuth card flows
   • embedding Web Chat components in React apps
   • webhook security verification headers
4. State management, dialog handling, and human handoff patterns90 min
   • UserState and ConversationState accessors
   • waterfall dialog step sequencing
   • escalation triggers to human support queues
   • storing conversation transcripts for auditing
   • session timeout and state cleanup
   • handling concurrent messages from the same user
5. Testing and debugging conversational flows locally60 min
   • Bot Framework Emulator configuration
   • inspecting bot activity payloads
   • mocking Azure OpenAI responses in tests
   • unit testing dialog state transitions
   • logging bot errors to Application Insights
   • troubleshooting channel connection issues
6. Optimizing bot responsiveness and latency60 min
   • typing indicator implementation
   • caching frequent bot responses
   • optimizing payload sizes for mobile channels
   • handling network timeouts gracefully
   • load testing bot endpoints
   • monitoring active user concurrency

Hands-on project: Develop a multi-turn conversational bot using the Bot Framework SDK connected to Azure OpenAI, capable of executing function calls and handing off to human agents.

You can automate infrastructure deployment and model rollouts using GitHub Actions and Azure DevOps pipelines.

1. Infrastructure as Code for AI services using Bicep and Terraform60 min
   • az deployment group create syntax
   • Terraform resource blocks for cognitive services
   • modularizing infrastructure definitions
   • state file management in Azure Storage
   • parameterizing environment configurations
   • dry-run plan execution
2. Building continuous integration workflows with GitHub Actions60 min
   • workflow trigger events (push, pull_request)
   • GitHub Secrets management for API keys
   • unit testing Python AI integration scripts
   • artifact publishing and retention
   • linting Bicep and Terraform code
   • parallel job execution setup
3. Automating continuous delivery and environment promotion60 min
   • environment approvals and gates
   • staging vs production deployment slots
   • blue-green deployments for AI microservices
   • automatic rollback triggers on failure
   • smoke testing endpoints post-deploy
   • notification integrations for pipeline status
4. Managing API keys and configuration settings via GitHub Secrets and Key Vault60 min
   • OIDC authentication for GitHub Actions to Azure
   • federated credentials configuration
   • secret rotation pipeline automation
   • secure environment variable injection
   • auditing secret access in pipelines
   • handling expired service principal credentials

Hands-on project: Write a multi-stage GitHub Actions YAML workflow that validates Bicep templates, runs automated test suites against Azure OpenAI endpoints, and deploys infrastructure.

You can configure comprehensive telemetry tracking, token usage alerts, and performance diagnostics in production.

1. Setting up distributed tracing and custom metrics in Application Insights60 min
   • telemetry client configuration in Python
   • tracking custom events and metrics
   • dependency tracking for external AI APIs
   • exception logging with stack traces
   • correlation IDs across distributed requests
   • sampling rate configuration for high-traffic apps
2. Monitoring token consumption and managing cost thresholds60 min
   • prompt token metrics extraction
   • completion token cost calculation
   • cost alert rule setup in Azure portal
   • quota threshold monitoring
   • attributing token usage to specific user tenants
   • detecting unusual usage spikes
3. Creating Kusto Query Language (KQL) dashboards for operational health60 min
   • summarize operator and aggregation functions
   • bin function for time-series grouping
   • render timechart and bar chart commands
   • percentile calculations for latency SLAs
   • filtering logs by error codes and severity
   • publishing shared workbooks to teams
4. Configuring alerts, action groups, and incident response runbooks60 min
   • metric alert rules on error rates
   • webhook integration for pager systems
   • email action groups for on-call engineers
   • runbook automation for auto-mitigation
   • testing alert trigger conditions
   • alert suppression during maintenance windows

Hands-on project: Set up an Azure Monitor workbook tracking latency percentiles, token consumption rates, and error frequencies for a deployed Azure OpenAI endpoint.

You can enforce data privacy, role-based access control, and content filtering aligned with enterprise compliance standards.

1. Implementing Role-Based Access Control (RBAC) and custom roles60 min
   • Cognitive Services OpenAI User role assignment
   • custom role definition JSON schema
   • principle of least privilege enforcement
   • subscription and resource group scope assignments
   • revoking stale permissions
   • auditing role assignments via Azure CLI
2. Enforcing data residency, privacy, and compliance frameworks60 min
   • regional data processing guarantees in Azure
   • customer-managed keys (CMK) configuration
   • data retention policies for AI telemetry
   • GDPR compliance and data subject requests
   • HIPAA compliance considerations for health data
   • evaluating data transfer agreements
3. Configuring Azure Policy and governance guardrails60 min
   • policy assignment creation and scoping
   • compliance auditing across subscriptions
   • denied resource types and SKUs
   • remediation tasks for non-compliant resources
   • initiative definitions grouping policies
   • exemptions management for special cases
4. Securing data transmission and managing encryption standards60 min
   • TLS 1.3 enforcement on endpoints
   • encryption at rest verification
   • secure storage account configurations
   • vulnerability assessments for storage blobs
   • network security group rule audits
   • certificate management for custom domains

Hands-on project: Configure Azure Policy definitions enforcing data residency constraints and restrict cognitive service access using Private DNS zones.

You can optimize cloud expenditure, scale inference endpoints dynamically, and handle high-volume traffic efficiently.

1. Analyzing and reducing Azure AI service costs60 min
   • Azure Cost Management tool navigation
   • reserved instances and commitment tiers
   • tier optimization S0 vs free tiers
   • unused resource cleanup scripts
   • cost attribution tagging strategies
   • forecasting monthly AI expenditure
2. Scaling inference workloads with managed endpoints and serverless options60 min
   • Azure Functions scaling triggers and limits
   • container instance resource allocation
   • provisioned throughput units allocation
   • load balancing strategies across regions
   • cold start mitigation techniques
   • scaling out search index replicas
3. Caching responses and optimizing prompt token efficiency60 min
   • Redis semantic caching implementation
   • response caching patterns for static queries
   • prompt token pruning and compression
   • payload optimization techniques
   • cache invalidation strategies for dynamic data
   • measuring cache hit ratio metrics
4. Handling high-concurrency traffic spikes without request dropping60 min
   • circuit breaker pattern implementation
   • queue-based load leveling with Azure Queue Storage
   • retry policies with jitter for API calls
   • throttling mitigation and client backpressure
   • graceful degradation of non-essential AI features
   • load testing methodology using Locust

Hands-on project: Implement an autoscale rule on an Azure Container Instance hosting a custom AI model based on CPU and request queue depth metrics.

You can architect, secure, test, and deploy a complete production-grade AI system on Azure meeting enterprise standards.

1. End-to-end architecture design for enterprise AI solutions120 min
   • Azure Well-Architected Framework pillars review
   • security baselines and threat modeling
   • cost estimation and budgeting models
   • scalability and disaster recovery planning
   • diagramming multi-service cloud architectures
   • presenting architectural trade-offs
2. Integrating multiple cognitive services into a cohesive pipeline120 min
   • Vision to Search data ingestion orchestration
   • speech-to-text voice bot pipeline wiring
   • document processing to database workflows
   • Event Grid event-driven triggers linking services
   • handling cross-service data transformations
   • end-to-end integration testing scripts
3. Production hardening, CI/CD deployment scripts, and monitoring120 min
   • GitHub Actions workflow finalization
   • Infrastructure as Code deployment validation
   • Application Insights alert rule verification
   • health check endpoints implementation
   • smoke testing live endpoints
   • final security audit and penetration check
4. Official AI-102 exam pattern analysis and timed mock test review120 min
   • case study questions breakdown and strategy
   • exam tips, traps, and common distractors
   • time management during the certification test
   • practice assessment review and remediation
   • booking and preparing for the Pearson VUE exam
   • final Q&A and career guidance wrap-up

Hands-on project: Deploy a complete multi-modal customer support portal that ingests scanned documents, indexes them in Azure AI Search, answers questions via Azure OpenAI, and logs telemetry to Application Insights.